arX

iv:1

009.

3265

v2 [

astr

o-ph

.CO

] 1

1 N

ov 2

010

Revised version November 11 2010 – Accepted for publication in ApJPreprint typeset using LATEX style emulateapj v. 11/10/09

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 OCCURS IN A SECULAR UNIVERSE:NO MAJOR MERGER-AGN CONNECTION⋆

Mauricio Cisternas1,20, Knud Jahnke1, Katherine J. Inskip1, Jeyhan Kartaltepe2, Anton M. Koekemoer3,Thorsten Lisker4, Aday R. Robaina1,5, Marco Scodeggio6, Kartik Sheth7,8, Jonathan R. Trump9, ReneAndrae1, Takamitsu Miyaji10,11, Elisabeta Lusso12, Marcella Brusa13, Peter Capak7, Nico Cappelluti13,

Francesca Civano14, Olivier Ilbert15, Chris D. Impey9 Alexie Leauthaud16, Simon J. Lilly17, Mara Salvato18,Nick Z. Scoville7, and Yoshi Taniguchi19

1 Max-Planck-Institut fur Astronomie, Konigstuhl 17, D-69117 Heidelberg, Germany2 National Optical Astronomy Observatory, 950 North Cherry Avenue, Tucson, AZ 85 721, USA

3 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA4 Astronomisches Rechen-Institut, Zentrum fur Astronomie der Universitat Heidelberg, Monchhofstr. 12-14, 69120 Heidelberg, Germany

5 Instituto de Ciencias del Cosmos (ICC), Universidad de Barcelona-IEEC, Martı i Franques 1, 08028 Barcelona, Spain6 IASF-INAF, Via Bassini 15, I-20133 Milano, Italy

7 California Institute of Technology, 1200 East California Boulevard, MC 249-17, Pasadena, CA 91125, USA8 Spitzer Space Center, California Institute of Technology, Pasadena, CA 91125, USA

9 Steward Observatory, University of Arizona, 933 North Cherry Avenue, Tucson, AZ 85721, USA10 Instituto de Astronomıa, Universidad Nacional Autonoma de Mexico, Ensenada, Mexico (PO Box 439027, San Diego, CA

92143-9027, USA)11 Center for Astrophysics and Space Sciences, University of California at San Diego, Code 0424, 9500 Gilman Drive, La Jolla, CA

92093, USA12 INAF-Osservatorio Astronomico di Bologna, Via Ranzani 1, I-40127 Bologna, Italy

13 Max-Planck-Institut fur Extraterrestrische Physik, Giessenbachstrasse 1, D-85748 Garching bei Munchen, Germany14 Harvard Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA

15 Laboratoire d’Astrophysique de Marseille, BP 8, Traverse du Siphon, 13376 Marseille Cedex 12, France16 LBNL & Berkeley Center for Cosmological Physics, University of California, CA 94720, USA

17 Department of Physics, ETH Zurich, CH-8093 Zurich, Switzerland18 Max-Planck-Institut fur Plasmaphysik, Boltzmanstrasse 2, D-85741 Garching, Germany and

19 Research Center for Space and Cosmic Evolution, Ehime University, Bunkyo-cho, Matsuyama 790-8577, Japan

Revised version November 11 2010 – Accepted for publication in ApJ

ABSTRACT

What is the relevance of major mergers and interactions as triggering mechanisms for active galacticnuclei (AGN) activity? To answer this long-standing question, we analyze 140 XMM-Newton-selectedAGN host galaxies and a matched control sample of 1264 inactive galaxies over z ∼ 0.3–1.0 andM∗ < 1011.7M⊙ with high-resolution HST/ACS imaging from the COSMOS field. The visual analysisof their morphologies by 10 independent human classifiers yields a measure of the fraction of distortedmorphologies in the AGN and control samples, i.e., quantifying the signature of recent mergers whichmight potentially be responsible for fueling/triggering the AGN. We find that (1) the vast majority(>85%) of the AGN host galaxies do not show strong distortions, and (2) there is no significantdifference in the distortion fractions between active and inactive galaxies. Our findings provide thebest direct evidence that, since z ∼ 1, the bulk of black hole (BH) accretion has not been triggeredby major galaxy mergers, therefore arguing that the alternative mechanisms, i.e., internal secularprocesses and minor interactions, are the leading triggers for the episodes of major BH growth. Wealso exclude an alternative interpretation of our results: a substantial time lag between merging andthe observability of the AGN phase could wash out the most significant merging signatures, explainingthe lack of enhancement of strong distortions on the AGN hosts. We show that this alternative scenariois unlikely due to: (1) recent major mergers being ruled out for the majority of sources due to thehigh fraction of disk-hosted AGN, (2) the lack of a significant X-ray signal in merging inactive galaxiesas a signature of a potential buried AGN, and (3) the low levels of soft X-ray obscuration for AGNhosted by interacting galaxies, in contrast to model predictions.Subject headings: galaxies: active — galaxies: evolution — galaxies: interactions — quasars: general

1. INTRODUCTION

cisternas@mpia.de⋆ Based on observations with the NASA/ESA Hubble Space

Telescope, obtained at the Space Telescope Science Institute,which is operated by AURA Inc, under NASA contract NAS5-26555; the XMM-Newton, an ESA science mission with in-struments and contributions directly funded by ESA MemberStates and NASA; European Southern Observatory under LargeProgram 175.A-0839; and the Subaru Telescope, which is oper-ated by the National Astronomical Observatory of Japan.

20 Member of the IMPRS for Astronomy and Cosmic Physicsat the University of Heidelberg

There is a general agreement that supermassive blackholes (BHs) lie at the centers of nearly all galaxies, orat least those with a bulge component. Additionally,strong correlations exist between the BH mass andvarious properties of the galactic bulge, including lumi-nosity (Kormendy & Richstone 1995; Magorrian et al.1998), stellar velocity dispersion (Gebhardt et al. 2000;Ferrarese & Merritt 2000; Tremaine et al. 2002), andstellar mass (Marconi & Hunt 2003; Haring & Rix2004). While it has been recently proposed thatthese correlations are just the product of a statistical

2 CISTERNAS ET AL.

convergence of several galaxy mergers over cosmictime (Peng 2007; Jahnke & Maccio 2010), these cor-relations have often been interpreted as the signatureof coupled evolution between the BH and its hostgalaxy (Kauffmann & Haehnelt 2000; Volonteri et al.2003; Wyithe & Loeb 2003; Granato et al. 2004;Hopkins et al. 2007; Somerville et al. 2008).Given that most galaxies are believed to have under-

gone a quasar phase, and that the central BH representsa relic of this event (Lynden-Bell 1967; Richstone et al.1998), the co-evolution picture is naturally very appeal-ing even while some aspects of it remain unclear. Ithas been suggested that most of the mass of the BH isbuilt up during the brightest periods of this quasar phase(Soltan 1982; Yu & Tremaine 2002). If there is such aconnection between the growth of the BH and its hostgalaxy, periods of quasar activity should occur alongsidethe growth of the bulge, and the mechanism that trig-gers the accretion onto a once quiescent BH, turning itinto an active galactic nucleus (AGN), should be tightlylinked with the overall evolution of the host galaxy. Thenature of AGN triggering is therefore of key importancefor our understanding of galaxy evolution in general.According to the current paradigm of hierarchical

structure formation, major mergers are a crucial elementin the assembly and growth of present-day galaxies (e.g.,Kauffmann et al. 1993; Cole et al. 2000; Somerville et al.2001; Bell et al. 2006; Jogee et al. 2009; Robaina et al.2010). A closer look into the behavior of simulated col-lisions between galaxies, beginning with the pioneeringwork of Toomre & Toomre (1972), suggests that grav-itational interactions are an efficient way of transport-ing material toward the very center of a galaxy. Merg-ers and strong interactions can induce substantial grav-itational torques on the gas content of a galaxy, de-priving it of its angular momentum, leading to inflowsand the buildup of huge reservoirs of gas in the cen-ter (Hernquist 1989; Barnes & Hernquist 1991, 1996;Mihos & Hernquist 1996; Springel et al. 2005; Cox et al.2006; Di Matteo et al. 2007; Cox et al. 2008).From early on, major mergers have been related to ob-

servations of powerful nuclear starbursts (Gunn 1979),and connections with quasar activity were made soonafter. Stockton (1982), in a study of luminous quasarswith close companions, suggested that these neighbor-ing galaxies could be survivors of a strong interactionwith the quasar. Further observational studies came tosupport this picture: more cases of quasars with closecompanions were found, and post-merger features weredetected in the host galaxies, whenever it was possibleto resolve them (e.g., Heckman et al. 1984; Gehren et al.1984; Hutchings et al. 1984, 1988; Stockton & Ridgway1991; Hutchings & Neff 1992). The merger–quasar con-nection scenario gained strength with the discovery of theultraluminous infrared galaxies (ULIRGs). More than95% of these were found in a merging state, some ofthem hosting an AGN. This led to the scenario in whichULIRGs and quasars were part of the same chain ofevents (Sanders et al. 1988a,b; Sanders & Mirabel 1996;Surace et al. 1998; Surace & Sanders 1999; Surace et al.2000; Canalizo & Stockton 2000, 2001).With the advent of the Hubble Space Telescope (HST ),

deep imaging of AGN host galaxies at higher red-shifts became possible with unprecedented resolution.

Many observational studies of luminous AGN found ahigh rate of merging signatures in their hosts and de-tected the presence of very close companions, whichbefore HST could not be resolved (e.g., Bahcall et al.1997; Canalizo & Stockton 2001; Zakamska et al. 2006;Urrutia et al. 2008). At the same time, deeper imag-ing of AGN host galaxies that were initially classifiedas undisturbed revealed post-merger features not previ-ously detected, both from space-based (Canalizo et al.2007; Bennert et al. 2008) and ground-based observa-tions (Ramos Almeida et al. 2010).There is, however, one major caveat for most of the

studies listed above: almost none of them made use of,or had the access to, an appropriate control sample of in-active galaxies; such a control sample is essential for dis-cerning if the merger rate is in fact enhanced with respectto the “background level”, i.e., the merger rate of inactivegalaxies. Only Dunlop et al. (2003) compared their sta-tistically complete sample of quasars against the quies-cent galaxy population, finding no difference in the struc-tural parameters between samples, as well as no enhance-ment in the large-scale disturbances. Even if not explic-itly, this showed a clear divergence from previous stud-ies regarding the merger-AGN connection scenario, andagreed with the very low frequency of post-merger sig-natures observed on Seyfert galaxies and low-luminosityAGN (Malkan et al. 1998; Schade et al. 2000).A new era of large HST programs now offers the po-

tential for resolving this discrepancy. The imaging oflarger, contiguous fields has yielded a large number ofobjects, making it possible to study AGN hosts at space-based resolution, and at the same time to compile a con-trol sample of non-active galaxies. Initial studies usingHST imaging by Sanchez et al. (2004) with the GalaxyEvolution from Morphologies and SEDs survey (GEMS,Rix et al. 2004) and by Grogin et al. (2005) with theGreat Observatories Origins Deep Survey (GOODS,Giavalisco et al. 2004) found no evidence for an enhance-ment in merging signatures of AGN hosts over controlgalaxy samples. If merger activity does not play a majorrole in AGN triggering, other methods to produce gas in-flows, build up the bulge, and fuel the BH should also beof importance. Alternate secular mechanisms—minor in-teractions, large scale bars, nuclear bars, colliding clouds,supernova explosions—can also lead to angular momen-tum removal and gas inflows from different scales to thecentral regions (for reviews, see Kormendy & Kennicutt2004; Wada 2004; Martini 2004; Jogee 2006). While theseprocesses have usually been related to Seyfert galax-ies and low-luminosity AGN (e.g., Simkin et al. 1980;Taniguchi 1999; Hopkins & Hernquist 2009), they couldpotentially play a larger role than usually reckoned formore luminous AGN as well. Although the results fromthe GEMS and GOODS surveys are highly intriguing,the field sizes of ∼0.22 deg2 and ∼0.08 deg2 respectivelywere still too small for definitive conclusions to be drawn.A suitably larger sample would be required to turn theseappealing hints into statements.In this context, we tackle this long-standing issue

by performing a comprehensive morphological analysisof a sample of X-ray-selected AGN host galaxies fromthe Cosmic Evolution Survey (COSMOS, Scoville et al.2007b), the largest contiguous area ever imaged with theHST (Scoville et al. 2007a; Koekemoer et al. 2007). Our

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 3

goal is to disentangle the actual relevance and predomi-nance of major galaxy mergers from the other suggestedmechanisms for the fueling of the BH.In the past, targeted high-resolution imaging of AGN

hosts has only been possible for small samples, while ex-tensive ground-based surveys with large samples havelacked of the necessary resolution to perform detailedmorphological studies at moderate redshifts. Earlier re-sults from the detailed analysis by Gabor et al. (2009),where the morphologies of ∼400 AGN host galaxy can-didates from the COSMOS field were parameterized,showed that these had an asymmetry distribution con-sistent with that of a control sample of inactive galax-ies, and lacked an excess of companions, already sug-gesting that major interactions were not predominantamong AGN as a triggering mechanism. Here we usethe largest sample of optically confirmed X-ray-selectedAGN ever imaged at HST resolution from the COSMOSsurvey and perform a visual inspection of the morpholo-gies of the host galaxies. We opt for a visual analysisof our galaxies over an automatic classification systembecause of the inherent problems and incompleteness ofthe latter in identifying mergers, even for some obviouscases, as cautioned by recent studies probing both meth-ods (Jogee et al. 2009; Kartaltepe et al. 2010). To estab-lish the relevance of our findings, we compare the AGNhosts to a matching sample of inactive galaxies from thesame exact data set.Throughout this paper we assume a flat cosmology

with H0 = 70 km s−1 Mpc−1, ΩM = 0.3, and ΩΛ =0.7. All magnitudes are given in the AB system unlessotherwise stated.

2. DATA SET AND SAMPLE

We will perform our analysis on a complete sampleof X-ray selected optically confirmed type-1 and type-2AGN from the COSMOS field.The COSMOS survey features the largest contiguous

area ever imaged with the HST. The location of the 1.64deg2 field, close to the celestial equator, allows accessfrom several major space and ground-based observato-ries, enabling a large multiwavelength coverage from X-ray to radio from supplementary observational projects.One of the most effective ways of finding AGN is to

make use of the X-ray emission due to the accretingBH (e.g., Mushotzky 2004). Complete coverage of thewhole COSMOS field in X-rays was achieved with theXMM-Newton (XMM-COSMOS, Hasinger et al. 2007;Cappelluti et al. 2009) through 55 pointings with a totalexposure time of ∼1.5 Ms. We use the catalog presentedin Brusa et al. (2010), which provides the most likelyoptical and infrared counterparts to the XMM sourcesbased on a likelihood ratio technique (see Brusa et al.2007 for details).From the X-ray catalog we draw a parent sample

of ∼550 sources classified as type-1 AGN from spec-troscopic surveys (Trump et al. 2007, 2009; Lilly et al.2007) revealing broad emission lines, and from spec-tral energy distribution (SED) fitting (Capak et al. 2007;Salvato et al. 2009; Ilbert et al. 2009). We also include asubsample of X-ray selected type-2 AGN, based on thoseused by Gabor et al. (2009) drawn from a parent sampleof ∼300 narrow emission line objects (Trump et al. 2007,2009).

Figure 1. The X-ray luminosity distribution of our sample in the2-10 keV energy band (solid line). For reference, we also show thedistribution of the type-1 AGN subsample only (dotted line).

In this paper, we analyze the morphological proper-ties of the AGN host galaxies. For this, we take ad-vantage of the high-resolution imaging of the COSMOSfield with the HST. These observations comprise 583orbits using the Advanced Camera for Surveys (ACS)with the F814W (broad I-band) filter (Koekemoer et al.2007). The imaging data feature an oversampled scale of0.′′03/pixel. Although the ACS survey of the COSMOSfield is highly homogeneous, the exact depth achieved isdependent on the angle of the telescope with the Sunat the time of the observations (Leauthaud et al. 2007).Ninety six out of the 575 pointings were made with anangle smaller than a critical value of 70, leading to aslightly shallower image. The limiting surface brightnesslevels above the background for the pointings made withan angle with the Sun larger and smaller than the criticalvalue are ∼23.3 mag arcsec−2 and ∼22.9 mag arcsec−2

respectively.We restrict our sample to the redshift range z ∼

0.3− 1.0. For the majority of our final sample, we usedhigh-confidence spectroscopic redshifts, while for the rest(20%), we used photometric redshifts by Salvato et al.(2009). The lower redshift cut is chosen due to the lownumber of AGN below z ∼ 0.3 (see Section 3 for fur-ther details), and also to avoid working with saturatedsources. The upper limit arises because the F814W filteris shifted into rest-frame UV for sources above z ∼ 1.This would mean that we would be specifically lookingat the light from young stars and star formation knots,and therefore at biased morphologies. At the same time,for the case of the type-1 AGN, the bright nucleus wouldstart to dominate due to its blue color, strongly outshin-ing the host and making it nearly impossible to resolve.Given our interest in the morphologies of our sample of

AGN host galaxies, we decided not to consider galaxiesfainter than IF814W = 24. Visual morphological classi-fication of these objects would be particularly difficult,and we determined that no consistent information couldbe extracted at this magnitude. For the case of the type-1 subsample, we applied this criterion after the nucleusremoval (see Appendix A for details).This selection yields 83 type-1 and 57 type-2 AGN. The

4 CISTERNAS ET AL.

median redshifts of the type-1 and type-2 subsamples are0.80 and 0.67 respectively. The type-2 subsample has amedian apparent IF814W of 20.9, slightly brighter thanthe type-1 subsample with 21.7 (after nucleus removal,see Appendix A) due to the higher median redshift of thelatter.Figure 1 shows the X-ray luminosity distribution of

our sources in the 2-10 keV energy band. The val-ues were obtained mainly from those calculated byLusso et al. (2010) and are complemented with those byMainieri et al. (2007). The median of our distributionlies at LX = 1043.5 erg s−1, which means that we areprobing a reasonably luminous representative AGN sam-ple. For reference, in Figure 1 we also show the X-rayluminosity distribution of the type-1 AGN subsampleonly, which dominates the overall distribution and hasa slightly brighter median LX (1043.6 erg s−1) than thetype-2 subsample (1043.3 erg s−1).

3. METHODOLOGY

In this paper we analyze the morphologies of a sampleof AGN host galaxies and of a control sample of inactivegalaxies using high-resolution HST/ACS single I-bandimages. In the following subsections we explain how webuilt the comparison sample (hereafter CS), the motiva-tion of choosing a visual inspection over an automaticmethod, and the classification scheme used.Analyzing the host galaxies of type-1 AGN is complex,

due to the presence of the bright active nucleus in theimages, that, depending on the contrast, can outshine thehost galaxy to different extents. We overcome this issuethrough a two-dimensional decomposition of the AGNand its host galaxy, modeling the bright active nucleuswith a point-spread function (PSF) and the host galaxywith a Sersic (1968) profile using GALFIT (Peng et al.2002, 2010), and then removing the nuclear contribution.This process is described in detail in Appendix A.

3.1. Comparison Sample

The large number of galaxies available from the COS-MOS HST observations provides us with the unique op-portunity of building a control sample from the samedata set that we draw our AGN from. For our study,we require the control sample to permit us elaborate acomparison regarding distortion features. On this re-spect the most relevant parameter is the signal-to-noiseratio (S/N), hence we construct the comparison sampleby selecting inactive galaxies matching each AGN bothin apparent IF814W magnitude and photometric redshift.This is both required and sufficient since (1) the S/N de-termines the visibility of the merger signatures, and (2)while the stellar masses might differ slightly (factors of∼2 at the same distance and brightness), the mass de-pendence on the merger rate is not strong, with only amodest increase for higher masses (Bundy et al. 2009).Specifically, for each AGN host galaxy we select 10 sim-

ilar comparison galaxies from the COSMOS ACS cata-log (Leauthaud et al. 2007). Each selected comparisongalaxy is required to have an IF814W magnitude withina range of IF814W = 0.1, and a photometric redshiftwithin a range of z = 0.05. If not enough galaxies werefound, the search ranges were increased by 10%. On av-erage, 1.8 iterations were performed to find the required

number of inactive galaxies for each AGN host. For thecase of the type-1 AGN subsample, the magnitudes ofthe host galaxies after the removal of the active nucleusare used for the selection of the control sample.With the inactive comparison sample in hand, we re-

move galaxies that are unlikely to be AGN host galaxycounterparts a priori via an initial visual inspection.Such galaxies include: (1) bulgeless disks and irregulars,which would represent a low-mass population, having nocorresponding partners on the AGN sample, and (2) forthe type-1 subsample, edge-on disks, which could in prin-ciple hold an AGN but this would be heavily obscuredand therefore not be a type-1.Finally, the construction of the control sample for the

type-1 AGN requires an additional effort. The nucleus re-moval process usually leaves residuals in the center whichcertainly affect any blind classification, making the type-1 AGN host galaxies readily discernible from the controlsample. To resolve this problem, we mock up our selectedinactive galaxies as AGN by adding a star in the centeras a fake active nucleus, as we describe in Appendix B.We then apply the same subtraction procedure as for theoriginal type-1 AGN, attempting to make the two sam-ples indistinguishable. As we show in Appendix C, anyeffects on the selection of the comparison sample due toflux variations caused by the nucleus subtraction processcan be neglected.Our final comparison sample consists of 1264 galaxies

in total. The IF814W and redshift distributions of the re-sulting type-1 and type-2 comparison samples are consis-tent with being drawn from the same parent distributionas the AGN subsamples, even after the removal of theunlikely counterparts described above. A Kolmogorov-Smirnov test on each couple of IF814W and redshift distri-butions confirms with probabilities > 38% that the AGNand control samples are consistent among each other (ingeneral < 5% is used to show that two distributions dif-fer).

3.2. Visual Classification

Merger events come in many different flavors due to thelarge parameter space involved (e.g., merger stage, view-ing angles, mass ratio, and gas fractions). Sometimesthey can be obvious at first sight, but some others canbe very subtle, or simply undetectable at the sensitiv-ity of the observations. At our redshift range and imageresolution, it has been shown that automatic classifica-tion methods to identify mergers tend to miss severalobvious cases, and cannot compete with visual inspec-tion (Jogee et al. 2009; Kartaltepe et al. 2010). On theother hand, when the numbers involved are over the tensof thousands, visual classification becomes impractical3

and an automatic approach would be needed. Generalmeasurements of structural parameters that can be cor-related with some physical process have proven to be agood compromise (e.g., Reichard et al. 2009, using thelopsidedness as a tracer of merging and star formation).Considering the above, in this paper we opt to identify

merger and interaction signatures visually. The numberof objects we are dealing with allows us to do so (∼1400in total), and the image quality deserves a detailed case-

3 With the notable exception of the citizen-based Galaxy Zooproject (Lintott et al. 2008, http://www.galaxyzoo.org).

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 5

Table 1Results from the Visual Analysis by the 10 Classifiers.

Classifier MC KI KJ JK AK TL AR MS KS JT µ

NAGN 140 140 140 140 22 40 57 140 38 98 ...Ntype−1 83 83 83 83 22 40 0 83 19 41 ...Ntype−2 57 57 57 57 0 0 57 57 19 57 ...NCS 1264 1264 1264 1264 177 357 537 1264 357 903 -Hubble typeBulgeAGN 25.7% 51.4% 31.4% 20.0% 40.9% 55.0% 29.8% 43.6% 26.3% 37.8% 35.2% ± 11.0%DiskAGN 74.3% 48.6% 68.6% 80.0% 59.1% 45.0% 70.2% 56.4% 73.7% 62.2% 64.8% ± 11.0%BulgeCS 24.6% 43.3% 29.6% 25.2% 47.5% 54.3% 29.4% 43.6% 17.9% 40.5% 34.3% ± 9.5%DiskCS 75.4% 56.7% 70.4% 74.8% 52.5% 45.7% 70.6% 56.4% 82.1% 59.5% 65.7% ± 9.5%DistortionsDist–0AGN 62.9% 43.6% 48.6% 56.4% 50.0% 47.5% 71.9% 56.4% 47.4% 55.1% 54.2% ± 7.5%Dist–0CS 65.5% 47.3% 60.1% 63.0% 67.8% 51.5% 78.0% 58.5% 51.5% 61.2% 59.9% ± 7.6%∆Dist−0 –2.6% –3.7% –11.6% –6.5% –17.8% –4.0% –6.1% –2.1% –4.2% –6.1% –5.6% ± 3.5%Dist–1AGN 24.3% 26.4% 45.0% 32.9% 40.9% 40.0% 21.1% 30.0% 50.0% 16.3% 30.8% ± 9.3%Dist–1CS 22.4% 28.3% 34.2% 26.9% 19.2% 34.2% 16.0% 33.5% 39.2% 17.8% 27.5% ± 6.5%∆Dist−1 1.9% –1.9% 10.8% 6.0% 21.7% 5.8% 5.0% –3.5% 10.8% –1.5% 3.2% ± 5.7%Dist–2AGN 12.9% 30.0% 6.4% 10.7% 9.1% 12.5% 7.0% 13.6% 2.6% 28.6% 15.0% ± 8.8%Dist–2CS 12.1% 24.4% 5.7% 10.1% 13.0% 14.3% 6.0% 7.9% 9.2% 20.9% 12.6% ± 6.5%∆Dist−2 0.8% 5.6% 0.7% 0.6% –3.9% –1.8% 1.1% 5.7% –6.6% 7.6% 2.4% ± 3.5%

Note. — We indicate the number of objects classified by each person for the AGN sample and individually for each type-1 and type-2 subsample,as well as for the comparison sample (CS). For the distortion classifications, we include the difference between samples as ∆Dist−X=Dist–XAGN –Dist–XCS . For each category, we include the mean, µ, and its dispersion, weighted according to the number of objects classified by each person.

Table 2Comparison of our mean Hubble type classification

with that of parametric estimators of galaxymorphologies

µa GALFITb ZESTc

BulgeAGN 35.2% 25.7% ...DiskAGN 64.8% 55.0% ...BulgeCS 34.3% 41.2% 19.6%DiskCS 65.7% 43.5% 67.8%

a Weighted mean of the 10 classifications (as in Table 1).b Percentages over 100% of the samples. The rest of theobjects had an intermediate Sersic index (with 2 ≤ n ≤ 3,see Appendix A for details).c Percentages over 100% of the comparison sample. Theremaining galaxies were classified as irregulars (11.3%) byZEST, and a few did not make it to the catalog (1.2%).

by-case examination.These visual studies can be subjective. In our case, the

absolute fraction of merging galaxies measured by visualclassifiers will depend on their own experience and back-ground, and hence it is plausible that they can differsubstantially among each other. Nevertheless, any per-sonal scale and criteria each classifier uses will be appliedequally on both samples, active and inactive galaxies.Therefore, a key quantity on our study will be, more thanthe absolute fractions of merging galaxies, the differencebetween the merging fractions measured by a given classi-fier. If we instead focus on how each individual classifierperceives one sample compared to the other, by consid-ering the differential between the merging fractions ofactive and inactive galaxies, this subjectiveness can beaccounted for. Furthermore, the consistency of this studyis improved by (1) using ten independent human classi-fiers to add statistical robustness and (2) mixing bothsamples of active and inactive galaxies so that the clas-sification is actually blind and therefore does not favoreither the AGN hosts or the inactive galaxies.We break the classification down into two parameters.

1. Hubble type. We attempt to state whether the host

galaxy belongs to one of the two basic morpholog-ical classes: bulge or disk dominated.

2. Distortion class. We define three classes regardingthe degree of distortion of the galaxy as follows.

(a) Dist-0. Galaxies that appear undisturbed,smooth and/or symmetric, showing no inter-action signatures. This also applies to caseswhere the small diameter of the galaxy doesnot allow a detailed analysis. We pay par-ticular attention to self-induced asymmetriessuch as dust lanes or star-forming regions,which are usually seen as small clumps in well-resolved spirals.

(b) Dist-1. Here we include objects with mild dis-tortions. This could be due to a minor mergerfor example, but at the same time could alsobe because of low S/N. This interaction classis a ”gray zone” in which most of the discrep-ancies in the classification between the 10 peo-ple arise.

(c) Dist-2. Strong distortions, potential signs forongoing or recent mergers. This class mainlyincludes galaxies which have highly disturbedmorphologies or show visible signatures ofstrong interactions, such as large tidal tails,arcs, debris, etc. Double-nucleus systems alsofall into this category.

Illustrative examples of the Hubble type and distortionclasses are shown in Figures 2 and 3, respectively.For the visual inspection, the classifiers had access to

FITS images which they could re-scale in order to lookfor high-contrast and subtle features that may have notshowed up at an arbitrary brightness scale.

4. RESULTS

The results from the visual classification by 10 people(MC, KI, KJ, JK, AK, TL, AR, MS, KS, and JT), for

6 CISTERNAS ET AL.

AGN host galaxies Inactive galaxies

Bulge-dom.

Disk-dom.

Figure 2. Example galaxy images arranged into different morphological classes with 100% agreement between the independent classifiers.The cutouts are 4.′′8 × 4.′′8. Black residuals at the center of some of the galaxies are residuals from the point source removal.

AGN host galaxies Inactive galaxies

Dist-0

Dist-1

Dist-2

Figure 3. Example galaxy images arranged into different distortion classes with 100% agreement between the independent classifiers.The cutouts are 4.′′8 × 4.′′8. Black residuals at the center of some of the galaxies are residuals from the point source removal.

both Hubble type and distortion classes, are shown inTable 1. For the different distortion classes, we show thedifference between samples (hereafter ∆) as the distor-tion fraction of the AGN minus that of the control sam-ple. The results are weighted according to the numberof objects classified by each person4 and used to calcu-late the mean fractions, µ, which we also display in Ta-ble 1. Figures 2 and 3 show examples of active and inac-tive galaxies which were classified with 100% agreement,arranged into the different Hubble type and distortionclasses respectively.

4 Each classifier looked at a minimum of ∼200 galaxies from thecombined sample. For each classifier the samples were shuffled, toassure that even if all of them decided to look at 200 galaxies, theywould be looking at different objects. On average, each galaxy wasclassified 6.3 ± 1.0 times.

4.1. Perception of the Hubble type

No morphology priors are applied in the selection ofour comparison sample, with the minor exception of thepruning of irregulars and edge-on disks as described inSection 3.1. In order to test whether the samples areconsistent regarding their morphological composition, wecompare the AGN and comparison samples in Table 1.Although the mean values show a high dispersion due tothe large discrepancies between classifiers, the results forthe AGN and comparison samples are in good generalagreement for each classifier.The large fraction of disks, in particular in the AGN

sample, is interesting. To verify if this could be due tosystematic bias by the classifiers, we will use two inde-pendent parametric estimators of the morphological typeavailable at hand. First we compare the results from our

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 7

Figure 4. Distributions of the difference in the Monte Carlo sampled distributions of Dist-2 fractions between the AGN and controlsamples for the ten classifiers. For each distribution, a deviation from zero difference (dotted line) towards positive values indicates a higherfraction of distorted active galaxies, whereas a deviation towards negative values shows a higher fraction of distorted inactive galaxies.

GALFIT models chosen earlier, which we extended toour type-2 subsample as well as to its comparison galax-ies. We identify sources as bulge- or disk-dominated ifthe best-fit results from GALFIT had Sersic indices ofn = 4 and n = 1 respectively. The rest of the galaxiesfell between the two. As a second test, we look up theresults for our comparison sample from the Zurich Es-timator of Structural Types (ZEST, see Scarlata et al.2007 for details), in which the structure of thousandsof COSMOS galaxies was quantified through a principalcomponent analysis over a combination of Sersic indexand five non-parametric diagnostics. The ZEST resultsshow the fractions of galaxies classified either as bulges ordisks. Of the remaining fraction classified as neither, themajority (11.3%) was classified as irregulars, most likelydue to the lack of sensitivity of these automatic classi-fication schemes to peculiar systems such as interactinggalaxies; this is consistent with the observed fraction ofhighly distorted comparison galaxies. Sixteen galaxies,accounting for the remaining 1.2%, did not make it intothe catalog.Table 2 shows both of these tests along with the

weighted mean fractions for comparison. It is clear thatthe numbers from these tests follow the trend seen in thevisual classification. These tests provide a lower limit tothe fraction of disks, with > 55% of our AGN samplebeing hosted by true disks.

4.2. The distortion fractions

Our prime interest lies in the observed difference in dis-tortion fractions between samples of active and inactivegalaxies. The absolute values in distortion fractions de-

termined by the 10 classifiers are of lesser interest sincethe internal calibration for the three distortion classesdiffers between the classifying individuals. Since anysubjectiveness will be applied equally to both active andinactive samples, using the differences in the distortionfractions instead of absolute levels removes the person-to-person calibration differences and allows an unbiasedinterpretation.Considering that the merging signatures we were look-

ing for could sometimes be faint and weak, we addressthe potential loss of sensitivity to such features due tothe slightly shallower limiting magnitudes for ∼17% ofthe pointings (i.e., those with Sun-angles of <70). Foreach person, we have also carefully analyzed the resultsby dividing their classified sample into sources with sun-angles either side of this critical angle. We find that thereis no statistically significant difference in the distortionfractions as a function of Sun-angle. In addition, as theassignment of individual objects to either a deep or shal-low field is effectively random, and given that the AGNdistortion fractions are compared directly with those ofa comparison sample selected from the same data set(and thus with the same limiting surface brightness is-sues), the overall impact on our results of any bias towardsmaller distortion fractions in the shallower fields wouldin any case be negligible.The objects that fell into the Dist-2 class were those

which presented the strongest distortions, and hence sig-natures of a major interaction, for each individual clas-sifier As any difference in recent major merger incidencewould show itself in this class, we will focus on the Dist-2results here.

8 CISTERNAS ET AL.

Figure 5. Combined posterior probability distribution of the dif-ference of highly distorted galaxies between the AGN and controlsample for the 10 classifiers. The central 68% confidence level ismarked with vertical dashed lines, which shows that the histogramis consistent with zero difference (dotted line), ruling out any sig-nificant enhancement of merging signatures on our sample of AGNhosts with respect to the comparison sample of inactive galaxies.

Figure 6. Cumulative distribution of the simulated Dist-2AGN

fractions, showing the 68% and 95% confidence levels with thedashed lines. As mentioned in the text, this confirms with a 95%confidence that the highly distorted AGN fraction can not be largerthan 24.08%.

4.2.1. Combining 10 classifications

In table 1 we have already listed the Dist-2 fractionsfor all classifiers, their mean values, and also the mean ofthe difference in Dist-2 fractions between the AGN andcomparison samples. This permits the following initialassessment under the assumption of Gaussian errors: thedifference (2.4%) is below the uncertainty of 3.5%, andhence it is not significant.Nevertheless, since the error distribution is in fact not

Gaussian but follows a binomial distribution (accordingto the number of distorted galaxies in a sample of givensize) it is important that we use the correct combinationof results in order to give answers to the two main ques-tions: (1) Is there a genuine difference between the frac-tions of strongly distorted AGN hosts and inactive galax-ies, and (2) with the given sample size, what differencein distortion fractions between samples can we actuallyrule out at a given confidence level—in this case we chose95%. The first question asks whether the given dataset

shows an enhanced AGN distortion fraction or not. Thesecond question probes the discriminative power of thissample, and allows us to gauge the actual importanceof a null-result in question (1), since a decreasing sam-ple size means an increasing uncertainty in the distortionfractions and hence small samples have near zero discrim-inative power.Using the correct binomial error statistics for the dis-

tortion fractions of AGN and inactive galaxies, we com-pute for each classifier the probability distribution for thedifference ∆Dist−2. This is done in the following way:(1) individially for each classifier, we Monte Carlo sam-ple their pair of Dist-2 binomial probability distributionsfor the AGN and comparison samples separately, (2) wecompute the difference between these randomly sampledvalues, (3) we repeat this process one million times foreach classifier which yields 10 distributions for ∆Dist−2,(4) we normalize these probability distributions by theDist-2CS values measured by each person as shown in Ta-ble 1, in this way removing any bias applied by each clas-sifier’s personal scale (Figure 4), and (5) now in “differ-ential” space, where we are insensitive to between-personscatter, we combine these 10 distributions by co-addingtheir histograms, weighted by the size of the sample eachperson classified5.The resulting probability distribution is shown in Fig-

ure 5. The histogram is fully consistent with zero differ-ence, as indicated by the central 68% confidence intervaldenoted by the vertical dashed lines estimated by the ar-eas at both ends, encompassing 16% each. This confirmsthe simple analysis from above: our study shows no sig-nificant difference between the fractions of strong distor-tions of AGN and inactive galaxies. Regarding the dis-criminative power of our sample, in Figure 6 we show thecumulative distribution of the Dist-2AGN fraction fromFigure 5. The distribution shows that with 95% confi-dence the distortion fraction of AGN is in any case notlarger than the inactive distortion fraction by a factorof 1.9, when considered relative to the mean distortionlevel found by the 10 classifiers (12.6%). Hence, the vastmajority of AGN host galaxies at z < 1 with the given lu-minosities do not show signatures of having experienceda recent major merger.

4.2.2. Mass dependency

Even if there is no overall difference between thefractions of highly distorted AGN and inactive galax-ies, it is still interesting to look at the situation inmass-space, and investigate the possibility that an en-hancement of the AGN merger fraction could be hid-den because we consider the sample as a whole, re-gardless of stellar mass. Major merging is a key ele-ment in the assembly and evolution of massive galax-ies (e.g., Bell et al. 2006; Lin et al. 2008; Bundy et al.2009; van der Wel et al. 2009; Robaina et al. 2010), andin order to test if the fraction of highly distorted AGNhost galaxies is significantly enhanced at the massive end(higher than ∼ 1010.5 M⊙), we have estimated stellarmasses for our samples of active and inactive galaxies.We use the calibration from Bell & de Jong (2001) basedon the Chabrier initial mass function. By obtaining the

5 This represents a combined Bayesian posterior probability dis-tribution with sample sizes as individual priors.

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 9

Figure 7. The combined differences in distortions of intermedi-ate (109.3 < M∗/M⊙ < 1010.5; top panel) and massive (1010.5 ≤

M∗/M⊙ < 1011.7; bottom panel) galaxies are shown. In bothcases, the central 68% confidence levels (dashed lines) are consis-tent with zero (dotted line).

V -band luminosities:

LV = 10−0.4(V−4.82) (1)

and assuming a common mass-to-light ratio from therest-frame (B−V ) color, we derive stellar masses in solarunits:

M∗ = 10−0.728+1.305(B−V ) × LV (2)

with all magnitudes in Vega zero point.For the inactive galaxies and the type-2 subsample we

obtain rest-frame B and V from the photometric catalogby Ilbert et al. (2010). For the type-1 AGN, however, wecannot use that information because it includes the con-tribution from the luminous AGN. Therefore, we obtainthe rest-frame V -band luminosities from the observedIF814W after the nucleus removal process and estimatethe color term by computing the linear regression overthe rest-frame (B−V ) colors as a function of redshift forthe type-2 AGN. This yields the relation

(B − V )V ega = 0.136 z + 0.541. (3)

The combined differences of highly distorted galax-ies for two bins of stellar mass (109.3 − −1010.5M⊙ and1010.5 − −1011.7M⊙) are shown in Figure 7. For bothsamples the ratio of galaxies occupying the massive binis roughly 2:1 relative to the less massive one, hence weare dealing with very massive galaxies. Even if for thegalaxies with stellar masses higher than ∼ 1010.5 M⊙

there is a modest enhancement in the distortion fractionof the AGN hosts over the control sample (Figure 7, bot-tom panel), it is again within the 68% confidence interval,i.e., it is not statistically significant. Therefore, it cannotbe considered as an empirical proof of an enhancementat the massive end.

5. DISCUSSION

From a detailed analysis of the results of our visualclassification we showed that the fractions of heavily dis-torted active and inactive galaxies are consistent withinthe central 68% confidence interval and that the Dist-2 fraction of AGN host galaxies is less than twice thatof the inactive galaxies at a 95% significance level, as

shown in Figures 5 and 6, respectively. Putting thesefindings in context, provided that the duration of mergersignatures and the visibility of the AGN phase overlapwith each other, this indicates that there is no evidencethat major merging plays a key role in the triggering ofAGN activity in our sample. But what about the pos-sible alternative scenario in which, in spite of a causalconnection between merging and AGN triggering, we donot detect an enhancement of merger signatures in theAGN population due to a significant time lag betweenthe interaction and the start of the AGN phase? Belowwe address this possible alternative interpretation withsome simple tests, and discuss the implications of ourresults.

5.1. Alternative Interpretation: Time Lag BetweenMerging and the Observability of the AGN Phase

Appealing simulations of mergers between gas-richgalaxies state that the peak of star formation and quasaractivity will occur during the final stages of the interac-tion, close to coalescence, within a more relaxed thandistorted bulge-like remnant (Di Matteo et al. 2005;Springel et al. 2005). In these models, during the firstpassage only modest starbursts are triggered and nomajor BH accretion occurs, and therefore the galax-ies would not be detected as AGN. Furthermore, adhoc models that include obscuration in galaxy mergers(Hopkins et al. 2005b) predict that, beginning from theearly stages of the interaction, the AGN is “buried” for∼90% of its lifetime by large column densities, only re-vealing itself toward the end of the merger. However, allthese models work with sub-grid prescriptions of BH ac-cretion and fail to spatially resolve the actual accretionprocess by several orders of magnitude.If there is indeed a substantial time lag after merg-

ing prior to the AGN activity becoming detectable, thenthe strong merging signatures we attempt to find couldhave already been washed out. Moreover, if AGN are ob-scured as the interacting galaxies coalesce, there could bea “contamination” population of undetected strong BHactivity occurring within our control galaxies undergo-ing a major merger. Finally, a third issue related to theobscuration plus time lag scenario is that the observed in-teractions that are occurring on a fraction (∼15%) of ourAGN host galaxies should be unrelated to the detectedBH accretion—under the assumption of a large time lagwe would not expect to see strong merging signatures.

5.1.1. AGN Hosted by Disks: Not a Relic from a MajorMerger

In the preceding text, we raised a possible alterna-tive explanation for our results, that most major merg-ers could be missed because the time lag between merg-ing and the observed AGN episode could be substantial,washing out the signatures that the HST/ACS resolutionallows us to detect.Models can provide us with some clues about the ob-

servability timescales during an interaction. For exam-ple, simulations of major mergers by Lotz et al. (2008)quantified that the strong signatures could still be de-tected 0.7 Gyr after the merger, by degrading their snap-shots to the resolution of HST z ∼ 1 imaging. Thus, inorder to explain the observed zero distortion enhance-ment, a lag of at least 0.7 Gyr between coalescence and

10 CISTERNAS ET AL.

the visible phase of the AGN would be required for allgalaxies6. It is, however, not straightforward to rely onthese studies to discard the time lag issue; given the largenumber of parameters involved in determining how longa merger signature will remain visible, it is plausible thatseveral late-stage mergers could have been missed. Al-though a merger between gas-rich galaxies can leave spec-tacular features for a long time, viewed from the wrongorientation they can be completely unnoticeable.While it is difficult to assess the relevance for the

timescale issue of major mergers being overlooked, wecan be reasonably confident that the remnant will notlook like a disk. Spheroidal and bulge-dominated galax-ies are usually said to be formed as a result of majormergers (e.g., Toomre 1977; Barnes & Hernquist 1996;Cox et al. 2006). However, it has also been statedthat disks can survive some major mergers, especiallyif the progenitors are gas-rich (e.g., Barnes & Hernquist1996; Springel & Hernquist 2005; Hopkins et al. 2009),nonetheless these kind of merger remnants have been ar-gued to not lead to a large bulge growth and significantBH fueling (Hopkins & Hernquist 2009). Likewise, it hasbeen argued that some gas-rich mergers can lead to theregrowth of the disk (Hopkins et al. 2009; Bundy et al.2010). Even so, the timescales involved for such a processcan be as much as an order of magnitude larger than thetypical quasar lifetime of 1-100 Myr (e.g., Porciani et al.2004; Hopkins et al. 2005b; Shen et al. 2007).For the significant fraction of AGN hosted by disks

found from our classification, we could safely say thatthe mechanism responsible for triggering those AGN wasnot a past major merger, suggesting also that since z ∼ 1alternative fueling methods seem to play a larger rolethan usually expected. Georgakakis et al. (2009), froma sample of X-ray-selected AGN, compared the lumi-nosity function of their disk-hosted AGN against theanalytic model of the X-ray AGN luminosity functionfor a stochastic accretion mode by Hopkins & Hernquist(2006). They showed that the model can reproduce theobservations, but at the same time the overall numberdensity of the observed disks was underpredicted, espe-cially at high X-ray luminosities. On our sample of 140AGN, 18 sources have LX ≥ 1044 erg s−1, from which 10of their host galaxies were classified as disk dominatedwith an agreement ≥80%. This suggests that alterna-tive BH fueling methods (i.e., those that do not destroythe disk) are not only more common on the overall AGNpopulation at z < 1, but also much more efficient thanthe existing models predict.

5.1.2. No veiled X-ray activity in merging galaxies

The aforementioned models leave the possibility thatwe could be missing an important fraction of AGN due togas and dust obscuration when a gas-rich major merger istaking place. Even though obscured AGN can still be de-tected through their hard X-ray emission (Hopkins et al.

6 For example, see Schawinski et al. (2010) who make an exten-sive case of the time lag scenario. They propose an all-merger-driven AGN phase with a time lag of ∼500 Myr for their sampleof early-type galaxies at z ∼ 0.05. Even if their result is mainlybased on the interpretation of their data as a causal sequence ofevents (and is subject to alternative explanations), they cautionthat their particular sample only accounts for a very small fraction(∼10%) of the overall AGN population found in the local universe.

2005a), it is possible that less luminous and highly ob-scured AGN lie below the detection threshold used tobuild the X-ray catalogs (Treister et al. 2004). The X-ray properties of such obscured objects have been suc-cessfully studied in the literature by the means of astacking analysis of X-ray data (e.g., Daddi et al. 2007;Fiore et al. 2009). If obscured AGN are being missed,they should be preferentially found in merging galaxies.Therefore, in order to test this scenario and search forthis potentially buried X-ray activity, we stack all theinactive galaxies regarded as highly distorted. Eighty-seven inactive galaxies fulfill our simultaneous criteriaof being individually classified as either Dist-1 or Dist-2with an agreement of ≥75%, and classified as Dist-2 withan agreement ≥ 65%.For this analysis, we take advantage of the higher sen-

sitivity of the Chandra observations of the COSMOSfield (C-COSMOS; Elvis et al. 2009), compared with theXMM-Newton data. Even though Chandra covered onlyhalf of the field (∼0.9 deg2), it has a flux limit three timesbelow the XMM-Newton sensitivity, which makes thetradeoff in smaller coverage absolutely justifiable, con-sidering that we want to detect possible X-ray sourcesbelow the XMM-COSMOS catalog sensitivity threshold.For the stacking of the X-ray data, we used the

CSTACK tool7 developed by one of the authors (T.M.),which includes a detailed bootstrapping error analysisthrough 500 realizations. Because the stacking is madefrommultiple observations, we consider the counts withina radius varying according to the off-axis angle, corre-sponding to 90% of the encircled counts. We stackedthe 45 objects that lie within the C-COSMOS area, af-ter excluding 1 object that was close to an X-ray source.We found an excess of soft 0.5–2 keV and hard 2–8 keVcount rates from the source region at modest levels of2.2σ and 2.4σ, respectively. Figure 8 shows the resultsof the stacking in the two energy bands, with the aver-age radii of 3.′′4 and 3.′′7 for comparison, within which nosource is noticeable above the background level.The lack of any obvious source after the stacking sug-

gests that this moderate excess could be in part due tothe expected emission from star-forming galaxies, andalso from extended source emission (e.g., from a galaxygroup). The possibility that a few sources dominate theoverall count rate is unlikely since (1) the shape of thecount rate distribution is that of a unimodal Gaussianand (2) no outliers are present. Therefore, it is doubt-ful that we are missing a significant fraction of accretingBHs hidden within the population of inactive galaxiesundergoing interactions.

5.1.3. No Enhanced Soft X-ray Absorption in Merging AGNHost Galaxies

As mentioned before, AGN obscuration due to the sur-rounding gas and dust during a major merger would af-fect mainly the soft X-ray energy band, while the hardband would remain unobscured. If we observe an AGNhosted by a merging galaxy, and this interaction was re-sponsible for the BH activity, we would expect to observea hard X-ray spectrum from this source. To trace theobscuration level of our interacting AGN host galaxies,we compute their X-ray hardness ratio (HR). The HR is

7 http://cstack.ucsd.edu

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 11

Figure 8. Stacked Chandra images of 45 inactive galaxies likelyto be undergoing a major interaction, on the soft 0.5-2 keV (left)and hard 2-8 keV (right) energy bands, showing the average radiiof the stacked sources as white circles. The cutouts are 12′′ × 12′′.

defined asHR = (H − S)/(H + S), (4)

where H and S stand for the hard (2–10 keV) and soft(0.5–2 keV) counts, respectively. At our redshift range,it is still safe to say that the HR values lower than –0.2correspond to an unabsorbed, soft spectrum (Hasinger2008).From our visual analysis, we have 13 AGN host galax-

ies regarded as highly distorted with high agreement ac-cording to the criteria used before. By computing the HRfor these objects, we find that, contrary to what modelspredict, all of these particular sources present soft X-rayspectra. All of them have HR values ≤–0.2, with a meanof –0.53, which shows a low attenuation in the soft band.It has been argued, however, that the HR diagnostic is

rather crude in terms of predicting obscuration, and in-deed, bright Compton-thick AGN can feature soft X-rayspectra due to photoionized gas (Levenson et al. 2006).Even so, this is only valid when the AGN is not observeddirectly, and we can easily establish that at least for thetype-1 subsample this would not the case, and that weare certainly looking at active, accreting BHs. Lookingonly at the seven type-1 objects from these likely merg-ing galaxies, we find that the average HR is –0.56 whichindeed suggests a low level of obscuration.One possible interpretation is that these interactions

are not related to the observed AGN episode and that areinstead only chance encounters. Dissipationless or gas-poor mergers could account for the lack of obscuration,but then it is unlikely that any strong merging signaturesand substantial accretion onto the central BH would takeplace directly due to these kind of events. Pierce et al.(2007) found the same result for X-ray-selected AGNhosted by interacting galaxies, suggesting that the ob-served interactions were not responsible for the fuelingof those accreting BHs.From another perspective, however, the models men-

tioned earlier are limited by the proposed pictureof the merger-ULIRG-feedback-quasar timeline (e.g.,Sanders et al. 1988a,b; Hopkins et al. 2008), which is al-ready regarded as oversimplified. The AGN phase is saidto happen after coalescence, but observations of largesamples of ULIRGs, all of them undergoing interactions,have found a significant scatter in the trends of AGNcontribution, accretion rate, and dust obscuration withmerging state (Veilleux et al. 2009). Some of these have

even been found to be dominated by the AGN in pre-merging state. Chaotic behavior during a merger eventcan lead to various unpredictable episodes of starburstand nuclear activity. Such episodic behavior can startmuch earlier than the final coalescence and can be re-sponsible for different periods of gas inflows, obscuration,and visibility, therefore explaining an already unobscuredmerger-induced AGN still early during the interaction,as traced by the soft spectra observed in our interact-ing active galaxies. This conclusion, at the same time,contradicts the alternative time lag scenario.

5.2. Major Merging: Not the Most Relevant Mechanism

Our analysis has demonstrated that the scenario inwhich mergers are responsible for triggering AGN after asignificant time lag is unlikely. The high fraction of disks,the lack of a hidden significant AGN signal in merginginactive galaxies, and the missing soft X-ray obscura-tion of interacting AGN hosts all appear to rule out thismodel as a possible explanation of our results. The ab-sence of any further evidence in support of this scenarioleads us to the only remaining possible interpretation ofour results: active galaxies are involved in major mergersno more frequently than inactive galaxies, and mergershave not played a leading role in AGN triggering for thelast 7.5 Gyr. Our results agree with the few recent stud-ies that have used a control sample (Dunlop et al. 2003;Grogin et al. 2003, 2005; Pierce et al. 2007; Gabor et al.2009; Reichard et al. 2009; Tal et al. 2009, and also withrecent results from the E-CDFS by Bohm et al. 2010,submitted), in the sense that the morphologies of theAGN host galaxies are not unusual and do not showa preference for merging systems. Of the studies men-tioned earlier which supported a merger-AGN connec-tion, many only provided circumstantial evidence forsuch a link, without any control sample comparisons.The lack of enhancement on merging signatures for

AGN hosts with respect to the background level indi-cates that there is no causal connection between mergingand AGN triggering up to z ∼ 1 and M∗ ∼ 1011.7M⊙,the galaxies dominating BH growth at these redshifts.It is still a plausible scenario that major mergers couldbe responsible for some of the brightest quasars; we donot intend to neglect this possibility, but in the con-text of a clean, large X-ray selected population of AGN,it is certainly not the most relevant mechanism. Thelarge fraction of AGN hosted by disk-dominated galaxiesshows that alternative mechanisms, i.e., stochastic pro-cesses and minor mergers dominate, for this sample ofobjects.The merger-starburst connection has also been widely

studied in the same perspective. Both mechanisms sharethe need for enough cold gas to be brought to the cen-tral regions of the galaxy, so it is worth mentioninganalogous conclusions from the recent literature: (1) in-deed, major mergers can trigger strong starbursts (e.g.,Mihos & Hernquist 1996; Springel 2000), but (2) notalways, as seen in models (Di Matteo et al. 2007) andobservations (Bergvall et al. 2003), and (3) its overallcontribution is relatively modest (Di Matteo et al. 2008;Jogee et al. 2009), with no more than 10% of star forma-tion in massive galaxies being triggered by major mergersat z ∼ 0.6 (Robaina et al. 2009).Different studies (e.g., Ballantyne et al. 2006; Hasinger

12 CISTERNAS ET AL.

2008; Li et al. 2010) have converged on proposing the fol-lowing scenario: the major merger-driven evolution dom-inates early in the universe, producing the bulk of thebrightest quasars at z = 2− 3. Around z ∼ 1 however, adifferent evolutionary mechanism takes over, with secularprocesses becoming the main triggers for the BH activ-ity and growth. While our analysis cannot be performedat higher redshifts with the current observational dataset, our results appear to fit this picture. Nevertheless,the overall relevance of major merging, even in the earlyuniverse, has yet to be determined. Other recent studiessuggest that secular processes play a much larger role:observations of massive star-forming galaxies at z ∼ 2have shown that their buildup has been dominated bycold rapid accretion and secular processes (Genzel et al.2008), without the need of major mergers. It has beenstated on the basis of dark matter simulations that thelikely number of major mergers is insufficient to accountfor the transformation of star-forming turbulent disks atz = 2 into ellipticals at z = 0 (Genel et al. 2008). Abroader view of the accretion history of dark matter ha-los by Genel et al. (2010), quantified that ∼60% of thedark matter in a given halo is contributed by mergers,with only ∼20% being major mergers. Instead, the rest(∼40%) of the dark matter would be accreted smoothly.This also agrees with recent work using smooth par-ticle hydrodynamic simulations, stating that galaxieshave acquired most of their baryonic mass through thecold mode of accretion (Keres et al. 2005, 2009). Fur-thermore, merger-free models have shown that isolatedgalaxies can reproduce the quasar duty cycles betweenz = 1 and 3 and feed their BHs with the recycled gasfrom evolving stars (Ciotti & Ostriker 2007) and evenreproduce the observed scaling relations (Lusso & Ciotti2010). Overall, these studies have shown that secularevolution can be highly relevant, also at the redshifts atwhich the peak of quasar activity occurs.

6. CONCLUSIONS

In this work, we performed a consistent visual analy-sis on the HST -based morphologies of a sample of 140X-ray-selected AGN host galaxies over z ∼ 0.3− 1.0 andM∗ < 1011.7M⊙, and compared them with a matchedcontrol sample of inactive galaxies under the same con-ditions. Our goal was to search for the presence of anysignificant connection between major merging and BHfueling as suggested by models and observational tests.In summary:

1. From our visual analysis, ∼85% of our AGN hostgalaxies show no strong distortions on their mor-phologies. Comparison with the control sampleshows that the distortion fractions are equal withinthe 68% central confidence level. Given our samplesize, we can state that at a 95% confidence levelthe highly distorted fraction of AGN hosts is less

than 1.9 times that of the inactive galaxies. Merg-ers and interactions involving AGN hosts are notdominant, and occur no more frequently than forinactive galaxies.

2. Over 55% of the AGN from our sample are hostedby disk-dominated galaxies, implying a triggeringmechanism that would not destroy the disk, i.e.,not a major merger. This also indicates that itis unlikely that we could be missing major merg-ers due to strong distortions having already beenwashed out over a large time lag prior to the ig-nition of the AGN. The presence of an impor-tant fraction of disk-dominated hosts on the AGNbrighter than LX > 1044erg s−1 suggests that sec-ular fueling mechanisms can be highly efficient aswell.

3. Through a detailed stacking analysis of the X-raydata of our inactive galaxies undergoing mergers,we did not find an underlying X-ray signal indi-cating the presence of a substantial population ofobscured AGN.

4. Looking at the hardness of the X-ray emission ofour AGN hosts that are clearly undergoing an inter-action, we found soft X-ray spectra in all of them,contradicting the expected obscuration in this bandpredicted by models. This can be either becausethe observed interactions are not responsible for theBH fueling or the unpredictable output of a mergerevent allows many accretion phases as well as anunobscured AGN, even during such early stages.

Our work explicitly suggests that, at least for the last7.5 Gyr, major merging has not been the most relevantmechanism in the triggering of typical AGN, and that thebulk of the BH accretion occurs through internal secu-lar processes and minor interactions. The alternativeinterpretation of a time lag between merger trigger andAGN onset is unlikely due to the zero enhancement of thedistortion fraction, the high incidence of disks, and theabsence of a significant X-ray signal in merging inactivegalaxies as a potential buried AGN population.

M.C. thanks G. De Rosa for productive discussions,C.M. Urry for useful comments, and the anonymous ref-eree for practical suggestions. M.C., K.J., and K.I. aresupported through the Emmy Noether Programme ofthe German Science Foundation (DFG). TM acknowl-edges support by CONACyT Apoyo 83564 and UNAM-DGAPA PAPIIT IN110209.Facilities: CXO, VLT:Melipal (VIMOS), HST

(ACS), Magellan:Baade (IMACS), Subaru (Suprime-Cam), XMM.

APPENDIX

A. AGN-HOST GALAXY DECOMPOSITION

The light distribution of the type-1 AGN is clearly dominated by the bright active nucleus, and because we wantto analyze the morphologies of their host galaxies, accurate removal of the nuclear source is of vital importance.This is done through a rigorous two-dimensional parametric fitting with GALFIT, with which we reduce each systemdown to a two-component model: a PSF to represent the AGN and a Sersic light profile accounting for the host

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 13

Figure 9. The difference in the observed magnitudes (IF814W ) of the comparison galaxies before (in) and after (out) the point sourceaddition/subtraction. The left-hand panel plots this difference against the initial magnitude, and the right-hand panel against the host tonucleus flux ratio, H/N. The 1σ deviation away from the mean is 0.23 mag, indicated by the shaded area centered at 0.03 mag.

galaxy. After subtraction of the modeled PSF, we are left with the host galaxy emission plus some residuals. Previoussimulations have shown that, at our resolution and S/N, it is sufficient with a single-component model to accountfor the host galaxy rather than a more complex, multi-component one (Sanchez et al. 2004; Simmons & Urry 2008).Regarding double-nucleus sources, which could need a dedicated modeling, we only found 1 object in our sample. Thisis also consistent with the visual inspection from Civano et al. (2010) on C-COSMOS sources. We checked for thisparticular source (Figure 3, bottom left), for which our method removes the brightest of the two nuclei. The otherpoint source is significantly fainter and does not dominate the overall galaxy brightness, hence not requiring a morecomplex decomposition.An appropriate initial guess of the parameters is recommended to get a faster and converging model with GALFIT.

We opt to run Source Extractor (Bertin & Arnouts 1996) on our cutouts to generate, in a fast and automatic way,rough estimates of the free parameters of the Sersic profile, such as coordinates within the image, observed magnitude,axis-ratio b/a, half-light radius Re, and position angle.To ensure a reliable decomposition, we take particular care that the host galaxy is modeled with the least possible

unnecessary flux transfer between PSF and the Sersic profile, which would result in either an over- or undersubtraction.We perform several GALFIT runs on each object with 3 different choices for the Sersic index n: we fix it to a n = 1exponential profile (Freeman 1970), to a n = 4 de Vaucouleurs (1948) profile, and we also leave it a as free parameterfor GALFIT to decide. To choose the right model, we need our best fit to be reasonable in terms of the resultingparameters. We require our host galaxy model: (1) not to be too concentrated or too shallow, meaning a half-lightradius between 2.5 pixels < Re < 100 pixels, (2) not to diverge to extreme elongations, therefore to have b/a > 0.5,and (3) to have its Sersic index within 0.5 < n < 8, for the free n case. We interpret that if the values run awayfrom these boundaries, GALFIT did not manage to model the underlying galaxy but instead could be accounting foruncertainties in the PSF.The model with the least χ2 is chosen between those that comply with the above criteria. If the model with the free

Sersic index is chosen among the 3 cases, we reassign the index and model it as follows: (1) if n < 2 then n = 1, (2) if2 ≤ n ≤ 3 then n = 2.5, and (3) if n > 3 then n = 4.A key aspect of the AGN-host galaxy decomposition is the choice of an accurate PSF, both for modeling the AGN

itself and for deconvolving the host galaxy light distribution. Even though the space-based HST provides extremelyprecise PSFs, instrumental effects are still important. The position of the target within the detector and the temporalvariability along different orbits can lead to discrepancies between the PSFs from the observations and the ones usedfor the analysis. This yields systematic errors in the image decomposition which can be critical for very bright AGN.The COSMOS survey provides us with the opportunity to minimize these spatial and temporal effects by using stellarPSFs from stars observed under the same conditions than our targets. For each object, we build specific PSFs byaveraging the nearest ∼30 stars in the same manner to other similar studies with large HST coverage (Jahnke et al.2004; Sanchez et al. 2004). The rms error from the creation of each PSF is propagated to the intrinsic variance of theAGN; the uncertainty of the object being fitted is required by GALFIT in order to converge to a minimum normalizedχ2.

B. CREATING MOCK AGN HOSTS

For our subsample of type-1 AGN, we build a special comparison sample of simulated AGN hosts by adding stars asfake nuclei to our inactive galaxies. To remain true to the characteristic blue colors of the AGN, we perform an initialselection of stars from the COSMOS ACS archive by placing color cuts in (B − V ) < 0.75 and (V − R) < 0.95. Foreach of the control galaxies, we look for stars that match the contrast level between the fluxes of the host and nucleus(H/N) of the corresponding AGN. With a matching star found, we simply add it over the centroid of the galaxy.We then apply the same point source removal procedure as for the original type-1 sources. PSFs are created exactly

as before, and the light contributions of the star and the underlying galaxy are separated using GALFIT. With theexception of 3 unsuccessful fits, we are left with 727 simulated nucleus-subtracted AGN host galaxies that will serveas an appropriate comparison sample for our type-1 AGN hosts.

14 CISTERNAS ET AL.

C. TESTING THE RELIABILITY OF THE IMAGE DECOMPOSITION

The creation of a sample of simulated nucleus-subtracted hosts from a starting point of real galaxies and stars givesus the opportunity to check the impact of our point source removal technique, and to see whether this technique isbiased. How significant are the residuals? We have performed photometry on the control galaxies before and after theaddition/subtraction of the fake nucleus. If a large magnitude offset were to be found, we would have had to considerreselecting our control sample, because we would inevitably be comparing active and inactive galaxies with differentobserved magnitudes. We find that, on average, the galaxies are fainter by 0.03 mag after the subtraction, with a 1σdeviation of 0.23 mag. Figure 9 shows the difference between the initial and recovered magnitudes for the hosts as afunction of the initial magnitudes and H/N ratio for our control galaxies. There is no obvious correlation between theoffset and the initial magnitudes of the galaxies, but as expected the recovered values tend to be less exact for morecompact galaxies and brighter active nuclei.These results show that this technique is trustworthy, and the offset found can be considered negligible and does not

affect our choice of a comparison sample.REFERENCES

Bahcall, J. N., Kirhakos, S., Saxe, D. H., & Schneider, D. P. 1997,ApJ, 479, 642

Ballantyne, D. R., Everett, J. E., & Murray, N. 2006, ApJ, 639,740

Barnes, J. E., & Hernquist, L. 1996, ApJ, 471, 115Barnes, J. E., & Hernquist, L. E. 1991, ApJ, 370, L65Bell, E. F., & de Jong, R. S. 2001, ApJ, 550, 212Bell, E. F., Phleps, S., Somerville, R. S., Wolf, C., Borch, A., &

Meisenheimer, K. 2006, ApJ, 652, 270Bennert, N., Canalizo, G., Jungwiert, B., Stockton, A., Schweizer,

F., Peng, C. Y., & Lacy, M. 2008, ApJ, 677, 846Bergvall, N., Laurikainen, E., & Aalto, S. 2003, A&A, 405, 31Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393Brusa, M., et al. 2007, ApJS, 172, 353—. 2010, ApJ, 716, 348Bundy, K., Fukugita, M., Ellis, R. S., Targett, T. A., Belli, S., &

Kodama, T. 2009, ApJ, 697, 1369Bundy, K., et al. 2010, ApJ, 719, 1969Canalizo, G., Bennert, N., Jungwiert, B., Stockton, A., Schweizer,

F., Lacy, M., & Peng, C. 2007, ApJ, 669, 801Canalizo, G., & Stockton, A. 2000, AJ, 120, 1750—. 2001, ApJ, 555, 719Capak, P., et al. 2007, ApJS, 172, 99Cappelluti, N., et al. 2009, A&A, 497, 635Ciotti, L., & Ostriker, J. P. 2007, ApJ, 665, 1038Civano, F., et al. 2010, ApJ, 717, 209Cole, S., Lacey, C. G., Baugh, C. M., & Frenk, C. S. 2000,

MNRAS, 319, 168Cox, T. J., Jonsson, P., Primack, J. R., & Somerville, R. S. 2006,

MNRAS, 373, 1013Cox, T. J., Jonsson, P., Somerville, R. S., Primack, J. R., &

Dekel, A. 2008, MNRAS, 384, 386Daddi, E., et al. 2007, ApJ, 670, 173de Vaucouleurs, G. 1948, Annales d’Astrophysique, 11, 247Di Matteo, P., Bournaud, F., Martig, M., Combes, F., Melchior,

A., & Semelin, B. 2008, A&A, 492, 31Di Matteo, P., Combes, F., Melchior, A., & Semelin, B. 2007,

A&A, 468, 61Di Matteo, T., Springel, V., & Hernquist, L. 2005, Nature, 433,

604Dunlop, J. S., McLure, R. J., Kukula, M. J., Baum, S. A., O’Dea,

C. P., & Hughes, D. H. 2003, MNRAS, 340, 1095Elvis, M., et al. 2009, ApJS, 184, 158Ferrarese, L., & Merritt, D. 2000, ApJ, 539, L9Fiore, F., et al. 2009, ApJ, 693, 447Freeman, K. C. 1970, ApJ, 160, 811Gabor, J. M., et al. 2009, ApJ, 691, 705Gebhardt, K., et al. 2000, ApJ, 539, L13Gehren, T., Fried, J., Wehinger, P. A., & Wyckoff, S. 1984, ApJ,

278, 11Genel, S., Bouche, N., Naab, T., Sternberg, A., & Genzel, R.

2010, ApJ, 719, 229Genel, S., et al. 2008, ApJ, 688, 789Genzel, R., et al. 2008, ApJ, 687, 59Georgakakis, A., et al. 2009, MNRAS, 397, 623Giavalisco, M., et al. 2004, ApJ, 600, L93Granato, G. L., De Zotti, G., Silva, L., Bressan, A., & Danese, L.

2004, ApJ, 600, 580

Grogin, N. A., et al. 2003, ApJ, 595, 685—. 2005, ApJ, 627, L97Gunn, J. E. 1979, in Active Galactic Nuclei, 213–225Haring, N., & Rix, H. 2004, ApJ, 604, L89Hasinger, G. 2008, A&A, 490, 905Hasinger, G., et al. 2007, ApJS, 172, 29Heckman, T. M., Bothun, G. D., Balick, B., & Smith, E. P. 1984,

AJ, 89, 958Hernquist, L. 1989, Nature, 340, 687Hopkins, P. F., Cox, T. J., Younger, J. D., & Hernquist, L. 2009,

ApJ, 691, 1168Hopkins, P. F., & Hernquist, L. 2006, ApJS, 166, 1—. 2009, ApJ, 694, 599Hopkins, P. F., Hernquist, L., Cox, T. J., Di Matteo, T., Martini,

P., Robertson, B., & Springel, V. 2005a, ApJ, 630, 705Hopkins, P. F., Hernquist, L., Cox, T. J., & Keres, D. 2008,

ApJS, 175, 356Hopkins, P. F., Hernquist, L., Cox, T. J., Robertson, B., &

Krause, E. 2007, ApJ, 669, 45Hopkins, P. F., Hernquist, L., Martini, P., Cox, T. J., Robertson,

B., Di Matteo, T., & Springel, V. 2005b, ApJ, 625, L71Hutchings, J. B., Crampton, D., Campbell, B., Duncan, D., &

Glendenning, B. 1984, ApJS, 55, 319Hutchings, J. B., Johnson, I., & Pyke, R. 1988, ApJS, 66, 361Hutchings, J. B., & Neff, S. G. 1992, AJ, 104, 1Ilbert, O., et al. 2009, ApJ, 690, 1236—. 2010, ApJ, 709, 644Jahnke, K., & Maccio, A. 2010, ApJ, submitted (arXiv:1006.0482)Jahnke, K., et al. 2004, ApJ, 614, 568Jogee, S. 2006, in Lecture Notes in Physics, Berlin Springer

Verlag, Vol. 693, Physics of Active Galactic Nuclei at all Scales,ed. D. Alloin, 143–+

Jogee, S., et al. 2009, ApJ, 697, 1971Kartaltepe, J. S., et al. 2010, ApJ, 721, 98Kauffmann, G., & Haehnelt, M. 2000, MNRAS, 311, 576Kauffmann, G., White, S. D. M., & Guiderdoni, B. 1993,

MNRAS, 264, 201Keres, D., Katz, N., Fardal, M., Dave, R., & Weinberg, D. H.

2009, MNRAS, 395, 160Keres, D., Katz, N., Weinberg, D. H., & Dave, R. 2005, MNRAS,

363, 2Koekemoer, A. M., et al. 2007, ApJS, 172, 196Kormendy, J., & Kennicutt, Jr., R. C. 2004, ARA&A, 42, 603Kormendy, J., & Richstone, D. 1995, ARA&A, 33, 581Leauthaud, A., et al. 2007, ApJS, 172, 219Levenson, N. A., Heckman, T. M., Krolik, J. H., Weaver, K. A.,

& Zycki, P. T. 2006, ApJ, 648, 111Li, Y., Wang, J., Yuan, Y., Hu, C., & Zhang, S. 2010, ApJ, 710,

878Lilly, S. J., et al. 2007, ApJS, 172, 70Lin, L., et al. 2008, ApJ, 681, 232Lintott, C. J., et al. 2008, MNRAS, 389, 1179Lotz, J. M., Jonsson, P., Cox, T. J., & Primack, J. R. 2008,

MNRAS, 391, 1137Lusso, E., & Ciotti, L. 2010, A&A, accepted (arXiv:1009.5292)Lusso, E., et al. 2010, A&A, 512, A34+Lynden-Bell, D. 1967, MNRAS, 136, 101Magorrian, J., et al. 1998, AJ, 115, 2285Mainieri, V., et al. 2007, ApJS, 172, 368Malkan, M. A., Gorjian, V., & Tam, R. 1998, ApJS, 117, 25

THE BULK OF THE BLACK HOLE GROWTH SINCE z ∼ 1 15

Marconi, A., & Hunt, L. K. 2003, ApJ, 589, L21Martini, P. 2004, in IAU Symposium, Vol. 222, The Interplay

Among Black Holes, Stars and ISM in Galactic Nuclei, ed.T. Storchi-Bergmann, L. C. Ho, & H. R. Schmitt, 235–241

Mihos, J. C., & Hernquist, L. 1996, ApJ, 464, 641Mushotzky, R. 2004, in Astrophysics and Space Science Library,

Vol. 308, Supermassive Black Holes in the Distant Universe, ed.A. J. Barger, 53–+

Peng, C. Y. 2007, ApJ, 671, 1098Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H. 2010, AJ, 139,

2097Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ,

124, 266Pierce, C. M., et al. 2007, ApJ, 660, L19Porciani, C., Magliocchetti, M., & Norberg, P. 2004, MNRAS,

355, 1010Ramos Almeida, C., Tadhunter, C. N., Inskip, K. J., Morganti,

R., Holt, J., & Dicken, D. 2010, MNRAS, 1609Reichard, T. A., Heckman, T. M., Rudnick, G., Brinchmann, J.,

Kauffmann, G., & Wild, V. 2009, ApJ, 691, 1005Richstone, D., et al. 1998, Nature, 395, A14+Rix, H., et al. 2004, ApJS, 152, 163Robaina, A. R., Bell, E. F., van der Wel, A., Somerville, R. S.,

Skelton, R. E., McIntosh, D. H., Meisenheimer, K., & Wolf, C.2010, ApJ, 719, 844

Robaina, A. R., et al. 2009, ApJ, 704, 324Salvato, M., et al. 2009, ApJ, 690, 1250Sanchez, S. F., et al. 2004, ApJ, 614, 586Sanders, D. B., & Mirabel, I. F. 1996, ARA&A, 34, 749Sanders, D. B., Soifer, B. T., Elias, J. H., Madore, B. F.,

Matthews, K., Neugebauer, G., & Scoville, N. Z. 1988a, ApJ,325, 74

Sanders, D. B., Soifer, B. T., Elias, J. H., Neugebauer, G., &Matthews, K. 1988b, ApJ, 328, L35

Scarlata, C., et al. 2007, ApJS, 172, 406Schade, D. J., Boyle, B. J., & Letawsky, M. 2000, MNRAS, 315,

498Schawinski, K., Dowlin, N., Thomas, D., Urry, C. M., &

Edmondson, E. 2010, ApJ, 714, L108Scoville, N., et al. 2007a, ApJS, 172, 38—. 2007b, ApJS, 172, 1

Sersic, J. L. 1968, Atlas de galaxias australes, ed. J. L. SersicShen, Y., et al. 2007, AJ, 133, 2222Simkin, S. M., Su, H. J., & Schwarz, M. P. 1980, ApJ, 237, 404Simmons, B. D., & Urry, C. M. 2008, ApJ, 683, 644Soltan, A. 1982, MNRAS, 200, 115Somerville, R. S., Hopkins, P. F., Cox, T. J., Robertson, B. E., &

Hernquist, L. 2008, MNRAS, 391, 481Somerville, R. S., Primack, J. R., & Faber, S. M. 2001, MNRAS,

320, 504Springel, V. 2000, MNRAS, 312, 859Springel, V., Di Matteo, T., & Hernquist, L. 2005, MNRAS, 361,

776Springel, V., & Hernquist, L. 2005, ApJ, 622, L9Stockton, A. 1982, ApJ, 257, 33Stockton, A., & Ridgway, S. E. 1991, AJ, 102, 488Surace, J. A., & Sanders, D. B. 1999, ApJ, 512, 162Surace, J. A., Sanders, D. B., & Evans, A. S. 2000, ApJ, 529, 170Surace, J. A., Sanders, D. B., Vacca, W. D., Veilleux, S., &

Mazzarella, J. M. 1998, ApJ, 492, 116Tal, T., van Dokkum, P. G., Nelan, J., & Bezanson, R. 2009, AJ,

138, 1417Taniguchi, Y. 1999, ApJ, 524, 65Toomre, A. 1977, in Evolution of Galaxies and Stellar

Populations, ed. B. M. Tinsley & R. B. Larson, 401Toomre, A., & Toomre, J. 1972, ApJ, 178, 623Treister, E., et al. 2004, ApJ, 616, 123Tremaine, S., et al. 2002, ApJ, 574, 740Trump, J. R., et al. 2007, ApJS, 172, 383—. 2009, ApJ, 696, 1195Urrutia, T., Lacy, M., & Becker, R. H. 2008, ApJ, 674, 80van der Wel, A., Rix, H., Holden, B. P., Bell, E. F., & Robaina,

A. R. 2009, ApJ, 706, L120Veilleux, S., et al. 2009, ApJS, 182, 628Volonteri, M., Haardt, F., & Madau, P. 2003, ApJ, 582, 559Wada, K. 2004, Coevolution of Black Holes and Galaxies, 186Wyithe, J. S. B., & Loeb, A. 2003, ApJ, 595, 614

Yu, Q., & Tremaine, S. 2002, MNRAS, 335, 965Zakamska, N. L., et al. 2006, AJ, 132, 1496